Openclaw Skill Development

Skill Development for OpenClaw

Overview

Skills are modular, self-contained packages that extend OpenClaw agent capabilities by providing specialized knowledge, workflows, and tools. Skills transform agents from general-purpose assistants into specialized experts equipped with procedural knowledge.

Key concepts:

• Skills live in ~/.openclaw/skills/<skill-name>/
• Each skill requires a SKILL.md file with YAML frontmatter
• OpenClaw uses metadata.openclaw block for platform-specific metadata
• Progressive disclosure: metadata always loaded, SKILL.md body on trigger, references as needed

Skill Anatomy

Directory Structure

~/.openclaw/skills/
└── skill-name/
    ├── SKILL.md              # Required: skill definition
    ├── references/           # Optional: detailed documentation
    ├── examples/             # Optional: working examples
    ├── scripts/              # Optional: utility scripts
    └── assets/               # Optional: output resources

SKILL.md Format

---
name: skill-name
description: "Describe when this skill should be used"
metadata:
  {
    "openclaw":
      {
        "emoji": "<emoji>"
      }
  }
---

# Skill Name

## Overview
Core principle in 1-2 sentences.

## When to Use
Bullet list with symptoms and use cases.

## Core Pattern
Implementation details and workflows.

## Quick Reference
Tables or bullets for scanning.

## Additional Resources
Pointers to references/, examples/, scripts/

Frontmatter Fields

name (required): Unique identifier for the skill

• Use kebab-case: my-skill-name
• Letters, numbers, hyphens only
• Max 1024 characters total for frontmatter

description (required): When to use this skill

• Use third person: "This skill should be used when..."
• Include specific trigger phrases
• Describe triggering conditions, NOT what the skill does
• Keep under 500 characters

metadata.openclaw (recommended): Platform-specific metadata

• emoji: Visual identifier for the skill in OpenClaw UI

Progressive Disclosure

Skills use a three-level loading system:

1. Metadata (name + description) — Always in context (~100 words)
2. SKILL.md body — When skill triggers (target 1,500-2,000 words)
3. Bundled resources — As needed by agents (unlimited)

What Goes Where

Skill Creation Process

Step 1: Understand the Skill

Identify concrete examples of how the skill will be used:

• What functionality should it support?
• What would a user say that should trigger it?
• Which agents in the roster would benefit most?

Step 2: Plan Resources

Analyze each example to determine:

• What scripts would be repeatedly rewritten? (-> scripts/)
• What documentation is needed while working? (-> references/)
• What templates or assets are needed? (-> assets/)

Step 3: Create Structure

mkdir -p ~/.openclaw/skills/skill-name/{references,examples,scripts}
touch ~/.openclaw/skills/skill-name/SKILL.md

Step 4: Write SKILL.md

Description best practices:

# GOOD: Third person, specific triggers, no workflow summary
description: "This skill should be used when the user asks to 'create a hook', 'add a scheduled task', or mentions cron job configuration."

# BAD: Summarizes workflow (agents may follow description instead of reading body)
description: "Use when executing plans - dispatches agent per task with review between tasks"

# BAD: Not third person
description: "Use this skill when working with hooks."

Body writing style:

• Use imperative/infinitive form (verb-first instructions)
• NOT second person ("You should...")
• Keep lean: 1,500-2,000 words
• Move detailed content to references/
• Reference all supporting files explicitly

Step 5: Validate

• SKILL.md has valid YAML frontmatter with name and description
• Description uses third person with specific trigger phrases
• Body uses imperative form, not second person
• Body is under 3,000 words (ideally 1,500-2,000)
• Detailed content moved to references/
• All referenced files exist
• metadata.openclaw block includes emoji

Step 6: Test and Iterate

Test the skill by simulating agent scenarios:

• Does it trigger on expected queries?
• Is the content helpful for intended tasks?
• Are there gaps in the instructions?

OpenClaw-Specific Considerations

Agent Roster Integration

Skills in OpenClaw serve 16 agents with different specializations. Consider:

• Which agents will use this skill most frequently?
• Does the skill need different guidance for different agent roles?
• Should the skill reference specific agents by name for routing?

Workspace Integration

Skills can reference OpenClaw workspace paths:

• ~/.openclaw/workspace/ for shared resources
• ~/.openclaw/workspace/agents-workspaces/<id>/ for agent-specific data
• ~/.openclaw/workspace/processes/PROCESSES.json for process definitions
• ~/.openclaw/workspace/standards/ for quality standards

Memory Integration

Skills can leverage OpenClaw's memory system:

• LanceDB at ~/.openclaw/memory/lancedb/ for semantic search
• SHARED_KNOWLEDGE.json for cross-agent knowledge
• Reference memory contents when relevant to skill domain

Common Mistakes

Weak Trigger Description

# BAD
description: "Provides guidance for working with automation."
# GOOD
description: "This skill should be used when the user asks to 'create a cron job', 'define a process', or 'automate agent tasks'."

Too Much in SKILL.md

Keep SKILL.md lean. Move detailed content to references/:

# BAD: 8,000 words all in SKILL.md
# GOOD: 1,800 words in SKILL.md + 2,500 in references/patterns.md

Missing Resource References

Always tell agents where to find additional information:

## Additional Resources
- **`references/patterns.md`** — Detailed patterns
- **`examples/working-example.sh`** — Complete example

Quick Reference

Minimal Skill

skill-name/
└── SKILL.md

Standard Skill (Recommended)

skill-name/
├── SKILL.md
├── references/
│   └── detailed-guide.md
└── examples/
    └── working-example.sh

Complete Skill

skill-name/
├── SKILL.md
├── references/
│   ├── patterns.md
│   └── advanced.md
├── examples/
│   ├── example1.sh
│   └── example2.json
└── scripts/
    └── validate.sh

Additional Resources

Reference Files

• references/skill-creator-original.md — Full original skill creation methodology